Google DeepMind CEO Demis Hassabis published an essay on July 14 titled "A Framework for Frontier AI and the Dawning of a New Age," calling for the United States to lead the creation of an independent, industry-funded body that would safety-test frontier AI models before they ship. The piece, published on his Substack and amplified on X, was covered the same day by TechCrunch, Axios and CNBC.

A FINRA for frontier AI

Hassabis's central proposal is a standards body modeled on FINRA, the private, industry-funded organization that regulates U.S. broker-dealers under SEC oversight. Under his framework, frontier labs would voluntarily share models with the body up to 30 days before public release so it could test for dangerous cyber, biological and "deception" capabilities. He envisions a majority-independent governing board that includes Turing Award-winning researchers alongside representatives from industry, government and the open-source community, and says he wants such a body stood up "before year end," according to Axios.

AGI is "a few years away"

The essay frames the proposal against a broader argument that artificial general intelligence — systems matching the full range of human cognitive capability — is probably only a few years off, and that its economic and social impact could arrive at roughly ten times the scale and speed of the Industrial Revolution. Hassabis describes the coming decade as offering "little margin for error" and casts the current moment as the "foothills" of a transition that will require international coordination on dual-use bio and cyber risks well before any single government could act alone.

Why it matters

Hassabis is the most prominent frontier-lab CEO yet to publicly back a specific, named regulatory model rather than general calls for "safety" or "alignment," and his FINRA comparison offers a concrete, self-funded structure that avoids handing direct authority to a single national regulator. Coming weeks after Anthropic and Google DeepMind leadership jointly called for a U.S.-led AI coalition at the G7, the essay signals that frontier labs are increasingly trying to shape binding oversight mechanisms proactively — before lawmakers in Washington or Brussels write rules for them. Whether an industry-seeded standards body can attract genuine independence and buy-in from competitors like OpenAI and Anthropic, rather than being seen as DeepMind setting the terms of its own oversight, is likely to be the first test of the proposal.